Target detection method and device, electronic equipment and storage medium

Abstract

The invention provides a target detection method and device, electronic equipment and a storage medium, a target detection model adopted in the method is obtained by pruning target channel groups in an alternative detection model, model parameters can be greatly reduced, and the detection accuracy is improved. Therefore, the operation time of the target detection model and the calculation amount of the embedded device carrying the target detection model can be reduced, the power consumption of the device is reduced, and the overall operation speed is improved. The alternative detection model comprises a plurality of channel groups, and the target channel group is a channel group meeting zero value invariance in the plurality of channel groups, so that the structure of the alternative detection model can be ensured to have sparsity; therefore, the target detection model obtained by pruning the target channel group still has the same performance as the alternative detection model, other operations do not need to be introduced to ensure that the performance of the models before and after pruning is consistent, and the cost required by the subsequent pruning can be reduced.

Description

technical field [0001] The present invention relates to the technical fields of computer vision and deep learning, in particular to a target detection method, device, electronic equipment and storage medium. Background technique [0002] In recent years, deep neural networks have been widely used in machine learning fields such as computer vision due to their good performance, such as image processing fields such as object detection and image classification, text processing fields such as semantic segmentation, and speech processing fields such as speech recognition. [0003] In order to improve the efficiency of deep neural network for target detection and reduce costs, a lightweight neural network model is introduced, that is, the unimportant feature channels in the neural network model are pruned, so that the model parameters are greatly reduced while still maintaining the original model. performance. The current main model pruning methods in the field of deep learning c...

AI Technical Summary

Problems solved by technology

[0004] Both of the above two types of schemes rely on a specific evaluation algorithm to evaluate the importance of a certain connection weight or a certain channel of convolution to the overall neural network model, and then iteratively perform the neural network model according to the priority of importance. pruning, which not only increases the time required for pruning, but also affects the performance of the neural network model

Images